Projects

Overview: EvoAgentX is an open-source framework designed to build, evaluate, and evolve LLM-based agents or agentic workflows in a self-evolving manner. It enables developers and researchers to create, test, and optimize AI agents through iterative feedback loops. Key features: - Agent Workflow Autoconstruction - Built-in Evaluation - Self-Evolution Engine - Plug-and-Play Compatibility - Comprehensive Built-in Tools - Memory Module - Human-in-the-Loop (HITL) Interactions Tech stack: Python Why it matters: EvoAgentX simplifies the process of building, evaluating, and evolving AI agents, making it easier to create dynamic and adaptable solutions for real-world applications.

Overview: TaylorWilsdon's Google Workspace MCP server integrates with all major Google Workspace services, offering comprehensive control through a single MCP server. It supports OAuth 2.1 multi-user authentication, three tiers of tools, and a full CLI for enhanced developer experience. Key features: - Full integration with Google Calendar, Drive, Gmail, Docs, Sheets, Slides, Forms, Tasks, Contacts, and Chat. - Stateless container deployment for easy setup. - Multi-user support across all services. - Extensive coverage of Google Workspace tools. - CLI and Code Mode for advanced developers. Tech stack: Python 3.10+. Why it matters: This MCP server is the most feature-complete solution available, offering unparalleled control over Google Workspace services, including multi-user support, rich fine-grained editing tools, and extensive coverage of all major integrations. It's also the only MCP server that can be hosted centrally and securely, making it ideal for organizations looking for a robust, managed solution.

### Overview Open CAD Studio is a cross-platform application designed for technical drawing, layout work, and solid modeling. It supports native DWG and DXF file formats, allowing users to open, edit, recover, and save these files directly without the need for conversion services. The application is built with Rust and includes features such as precise 2D drafting, documentation tools, kernel-backed 3D modeling, GPU-accelerated rendering, and extensible workflows. ### Key Features - **Native Drawing Workflow** – Open, edit, recover, and save DWG and DXF files without a conversion service. - **Precise 2D Drafting** – Lines, polylines, curves, splines, hatches, object snaps, tracking, layers, blocks, and external references. - **Documentation Tools** – Text, dimensions, leaders, tolerances, tables, model space, paper space, viewports, and plot styles. - **Kernel-Backed 3D Modeling** – Solid primitives, extrusion, revolution, sweep, loft, Boolean operations, and ACIS entity tessellation. - **GPU Rendering** – Accelerated 2D and 3D viewports through `wgpu`, with orthographic and perspective cameras. - **Extensible Workflows** – Native plugins, command scripts, headless conversion, and a line-based JSON automation API. ### Tech Stack Rust is the primary programming language used to build Open CAD Studio, leveraging its strengths in performance, security, and reliability. ### Why It Matters Open CAD Studio is designed for

### Overview: II-Agent is an open-source AI agent framework designed for real-world applications, built with the Apache-2.0 license. It offers a range of features including mobile app development, website app creation, interactive website generation, and more, with a focus on flexibility and cost control. ### Key Features: - **Build**: Supports mobile app development, website app development, storybook generation, video & image generation, live editing, and project planning. - **Research**: Provides fast and deep research capabilities, including interactive website generation. - **Automate & Integrate**: Features built-in and custom skills, app integrations, and faster execution compared to previous versions. - **Everything Else**: Includes chat capabilities, agent general tasks, document editing, and slide creation. ### Tech Stack: Python, Docker, uv, Node.js, and various LLM providers. ### Why it Matters: II-Agent simplifies the process of building intelligent agents, making it accessible to developers, research teams, and enterprises alike. Its open-source nature and flexible architecture allow for customization and cost control, making it a valuable tool for anyone looking to integrate AI into their projects.

Overview: TensorFlow.NET is a .NET Standard binding for Google's TensorFlow, allowing C# and F# developers to develop, train, and deploy Machine Learning models. Key features: - Cross-platform .NET Standard binding for TensorFlow - Built-in Keras high-level interface - Released as an independent package TensorFlow.Keras Tech stack: C# Why it matters: TensorFlow.NET simplifies the implementation of TensorFlow models in C# and F#, enabling .NET developers to leverage the full power of TensorFlow without the need for C++ knowledge.

Overview: Catalyst is a PyTorch framework designed for deep learning research and development, emphasizing reproducibility, rapid experimentation, and codebase reuse. It aims to help researchers and developers create novel models rather than reinvent train loops. Key features: - Catalyst is a PyTorch framework. - It focuses on reproducibility and rapid experimentation. - It encourages the reuse of code and models. - It supports multiple operating systems, including Linux, macOS, and Windows Subsystem for Linux (WSL). Tech stack: Python Why it matters: Catalyst simplifies deep learning research by providing a streamlined environment for experimentation and model development. It ensures that researchers can focus on innovation rather than the complexities of traditional deep learning frameworks. The framework's design emphasizes ease of use, making it accessible to both beginners and experienced developers. Repository: catalyst-team/catalyst GitHub stars: 3388 Primary language: Python GitHub description: Accelerated deep learning R&D

Overview: OP Vault is a project that utilizes the OpenAI and Pinecone Vector Database Stack to provide users with long-term memory capabilities through the integration of their own custom knowledge bases. This tool allows for the upload of various document types, enabling users to ask OpenAI questions about the specific content of their uploaded documents, providing accurate and relevant answers based on the provided files. Key features: - Upload custom knowledge base files (PDF, txt, epub, etc.) - Retrieve answers based on the content of uploaded documents - See filenames and specific context snippets informing the answers - Explore the power of the OP Stack (OpenAI + Pinecone Vector Database) - Load entire libraries' worth of books into the vault Tech stack: JavaScript, Golang, Node.js, Go, Poppler Why it matters: OP Vault is a practical and valuable tool for knowledge extraction and question-answering, making it useful for extracting information from various documents and providing context-sensitive answers.

### Overview: NVIDIA PhysicsNeMo is an open-source PyTorch framework designed for building, training, and fine-tuning deep learning models in the field of physics machine learning, scientific machine learning, and AI for science and engineering. It provides reusable library components and end-to-end training recipes. ### Key Features: - **Physics Machine Learning (Physics ML)**: Enables the development of physics-based machine learning models. - **Scientific Machine Learning (SciML)**: Facilitates the training of models using scientific data and simulations. - **AI for Science and Engineering**: Integrates AI techniques to solve complex problems in physics and engineering. ### Tech Stack: Python ### Why It Matters: NVIDIA PhysicsNeMo is essential for researchers and engineers looking to leverage the power of deep learning and physics-based modeling for advanced simulations and predictive analytics in fields such as aerospace, meteorology, and semiconductor engineering. Its open-source nature and extensive documentation make it accessible and valuable for both academic and industrial applications. Repository: NVIDIA/physicsnemo GitHub Stars: 3371 Primary Language: Python Installation Options: Refer to [installation options](#installation-options) for CUDA 12, a basic install, optional features, or source setup.

Overview: Agent of Empires (AoE) is a session manager for AI coding agents, providing easy access to multiple Claude Code, OpenCode agents via TUI, web, CLI, and HTTP API surfaces. It supports persistent tmux sessions, Git worktrees, multi-repo workspaces, and Docker, Podman, and Apple Containers sandboxing. Key features: - TUI, web, CLI, and HTTP API surfaces - Status detection, notifications, and persistent tmux sessions - Git worktrees and multi-repo workspaces - Docker, Podman, and Apple Containers sandboxing - Structured and terminal views, including mobile access - Diff review, profiles, repo hooks, and custom agent commands Tech stack: Rust Why it matters: Agent of Empires simplifies managing multiple AI coding agents, making it easier to work with complex projects and collaborate with others.

Overview: `audioFlux` is a comprehensive library for audio and music analysis, featuring advanced feature extraction methods and deep learning support for various tasks such as classification, separation, and music information retrieval. Key features: - Systematic and multi-dimensional feature extraction and combination for flexible research and analysis. - High-performance, C++ implementation with FFT hardware acceleration for efficient large-scale data feature extraction. - Supports mobile end with real-time audio stream calculation capabilities. Tech stack: C Why it matters: `audioFlux` simplifies audio analysis tasks, enabling researchers and developers to easily integrate deep learning models for audio classification, separation, and more, making it a valuable tool in the audio field. Repository: libAudioFlux/audioFlux GitHub stars: 3372 GitHub description: A library for audio and music analysis, feature extraction.

### Overview: Sepia is a powerful AI writing skill designed for any Agent Skill-compatible agent, offering a native plugin for Claude Code, Codex, Grok Build, and Antigravity. It repairs the narrative architecture of fiction before any word choice is altered, ensuring professional documents receive tailored rules for their specific venues. The skill is based on the research from StoryScope, a groundbreaking study in AI fiction detection. ### Key Features: - **Narrative Architecture Repair**: Sepia corrects the narrative architecture of fiction, eliminating unnecessary explanations of themes, simplifying causal chains, and mixing emotional modes. - **Sentence Rhythm and Chinese Calibration**: It adjusts the sentence rhythm and calibrates Chinese writing to ensure optimal readability and coherence. - **Native Plugins**: Supports native plugins for Claude Code, Codex, Grok Build, and Antigravity, making it highly versatile and compatible with a wide range of agents. ### Tech Stack: Python ### Why It Matters: Sepia is a significant advancement in AI writing skills, particularly for humanizing AI-generated content. By focusing on the narrative architecture and sentence structure, Sepia ensures that the AI-generated text is more aligned with human-written prose, improving its readability and coherence. This makes it particularly valuable for professional documents and fiction, where maintaining a consistent narrative structure is crucial. ### Repository: - **GitHub Repository**: [Nanako0129/sepia](https://github.com/Nanako0129/sepia) - **Stars**: 3,277 - **Description**: De-AI

Overview: LangGraphJS is a TypeScript framework designed to build resilient language agents as graph-based models. Key features: - Graph-based model architecture - Easy integration with LangChain ecosystem - Scalable and efficient language processing Tech stack: TypeScript Why it matters: LangGraphJS simplifies the development of language agents by leveraging graph-based models, making it easier to create robust and scalable applications.

### Overview: An open-source research agent system for your Zotero library, integrating Large Language Models to enhance your research workflow. ### Key features: - Chat with current PDFs, selected text, figures, screenshots, and uploaded documents directly inside Zotero. - Get grounded answers with citations that jump back to the source passage. - Compare multiple open papers or add external files as extra context. - Save answers, full conversations, and research notes to Zotero notes or local Markdown folders such as Obsidian and Logseq. - Enable Agent Mode for library-wide read, search, tagging, metadata, import, note-editing, and organization workflows. - Search the current public web and read relevant pages with Tavily, with source links attached to the answer. - Use your preferred backend: API keys, local models, ChatGPT WebChat, Codex App Server, or Claude Code. ### Tech stack: TypeScript ### Why it matters: Enhance your research process by integrating Large Language Models directly into your Zotero library, making it easier to ask questions, summarize papers, compare sources, and save notes without leaving your library.

Overview: Torch-Pruning is a comprehensive framework for structural pruning, enabling efficient model compression with a focus on structural pruning. It supports a wide range of deep neural network architectures, including large language models, vision foundation models, and more. Key features: - Supports pruning of models from various libraries (Huggingface, Timm, Torchvision, etc.). - Deploys the **DepGraph** algorithm to group and remove coupled parameters for efficient pruning. - Enables structural pruning for a variety of models, including LLMs, SAM, Diffusion Models, Vision Transformers, ConvNext, Yolov7, SSD, ResNe(X)t, DenseNet, RegNet, DeepLab, and more. Tech stack: Python Why it matters: Torch-Pruning simplifies the process of pruning deep neural networks, making it easier to compress models without losing important features, and it supports a broad range of architectures, making it a valuable tool for researchers and practitioners alike.

### Overview: This repository collects and categorizes papers on diffusion models, providing a comprehensive survey and taxonomy. The survey paper, "Diffusion Models: A Comprehensive Survey of Methods and Applications," has been accepted by the ACM Computing Surveys journal. The repository aims to keep up with the rapid development of diffusion models by updating both the arXiv paper and the repository. ### Key Features: - **Algorithm Taxonomy**: Categorizes diffusion models based on their sampling and likelihood optimization techniques. - **Sampling-Acceleration Enhancement**: Includes methods like SDE Solver, ODE Solver, Optimized Discretization, Knowledge Distillation, and Truncated Diffusion. - **Likelihood-Maximization Enhancement**: Covers Noise Schedule Optimization, Reverse Variance Learning, and Exact Likelihood Computation. - **Data with Special Structures**: Focuses on data with manifold and invariant structures, discrete data, and data with special structures. - **Diffusion with (Multimodal) LLM**: Discusses simple combinations, deep collaboration, and applications in computer vision, natural language processing, temporal data modeling, and more. - **Diffusion with DPO/RLHF**: Explores methods for data purification and generating synthetic data for robust learning. - **Connections with Other Generative Models**: Connects diffusion models with variational autoencoders, generative adversarial networks, normalizing flows, autoregressive models, and energy-based models. ### Tech Stack: The repository uses Markdown and GitHub Pages for its structure and content. ### Why it Matters: This repository serves as a valuable resource for

### Overview: VideoChatGPT is a state-of-the-art video understanding model that integrates advanced video and image chatting capabilities with the capabilities of the popular ChatGPT model. This project leverages the power of both Qwen and GPT-3 to create a highly versatile and efficient video chatbot. ### Key Features: - **Video and Image Chatting**: Seamlessly integrate video and image understanding with ChatGPT for real-time conversations. - **Model Performance**: Achieves the best performances among open-source VideoLLMs on MLVU benchmark. - **Efficient Inference**: Speeds up the inference process of VideoChat2 using VLLM. ### Tech Stack: Python ### Why It Matters: VideoChatGPT is a significant advancement in the field of video understanding and chatbot technology, providing a powerful tool for real-time video and image chatting. Its ability to integrate with the capabilities of ChatGPT makes it highly versatile and efficient, making it a valuable addition to the developer community. ### Repository: Repository: OpenGVLab/Ask-Anything GitHub Stars: 3360 GitHub Description: CVPR2024 Highlight [VideoChatGPT] ChatGPT with video understanding! And many more supported LMs such as miniGPT4, StableLM, and MOSS. ### Primary Language: Python ### Updates: - **2026/07/17**: Release VideoChat3, a fully open, efficient 4B Video MLLM for general, long-form, and streaming video understanding. - **

### Overview: OpenOutreach is an open-source AI agent designed for B2B lead generation. It discovers the people who fit your product and target market, explains why each one does, and emails them directly from your mailbox. Self-hosted and CLI-based, it requires a single install and one command to get started. ### Key Features: - **Self-Hosted CLI**: One install, one command. - **Self-Hosted**: No platform terms of service. - **No Cold-Email Sequencing**: No list upload required. - **No Lead Database**: Output is per person, plain language. - **No Scrape**: No scraping or account management. - **No Account**: No banned account risk. ### Tech Stack: Python ### Why It Matters: - **Self-Hosted**: No need for a separate server. - **One Install**: No need to set up multiple installations. - **One Command**: Simple, no additional steps. - **Zero Platform Terms**: No terms of service to agree to. - **Zero Scrape**: No scraping or scraping-related issues. - **Zero Account Risk**: No risk of getting banned. ### Repository: - **GitHub Repository**: eracle/OpenOutreach - **Stars**: 3344 - **Description**: Open-source AI agent for B2B lead generation — finds the people who fit, explains why each one does, and emails them from your mailbox. Self-hosted CLI, one install. ### Demo:  ### Quick Start: ```bash uv tool

### Overview: Whisper ASR Box is a comprehensive speech recognition toolkit that leverages OpenAI's Whisper models for high-quality speech recognition, multilingual support, and advanced features like voice activity detection and speaker diarization. This project aims to provide a robust and versatile solution for developers and enthusiasts alike. ### Key Features: - **Multiple ASR Engines Support**: Supports OpenAI Whisper, Faster Whisper, and WhisperX models. - **Multiple Output Formats**: Outputs in text, JSON, VTT, SRT, and TSV formats. - **Word-Level Timestamps**: Provides precise timestamps for each word in the recognized text. - **Voice Activity Detection (VAD)**: Filters out silent periods in the audio. - **Speaker Diarization**: Utilizes WhisperX for better speaker identification. - **FFmpeg Integration**: Compatible with various audio/video formats for seamless integration. - **GPU Acceleration**: Supports GPU acceleration for faster processing times. - **Model Loading/Unloading**: Configurable to load and unload models at will. - **REST API**: Provides a Swagger documentation for easy integration with other systems. ### Tech Stack: - Python - Docker - FFmpeg ### Why It Matters: Whisper ASR Box is designed to be a versatile tool for developers working on speech recognition projects. Its ability to handle multiple models and output formats makes it highly adaptable to different use cases. The inclusion of FFmpeg integration ensures compatibility with a wide range of audio and video formats, making it a valuable asset for developers working with diverse media content. ### Repository: The repository is maintained

### Overview: The Local File Organizer is an AI-powered tool designed to organize local files on your device, ensuring privacy and seamless access. Leveraging advanced models like Llama3.2 3B and Llava v1.6, it scans, restructures, and organizes files for quick access and easy retrieval. ### Key Features: - **AI-Powered File Organization:** Automatically categorizes and organizes files based on content and visual elements. - **Privacy-Respecting:** All AI processing happens on your local device, ensuring no data leaves your computer. - **Versatile File Types:** Supports text files, spreadsheets, images, and more. - **User-Defined Sorting:** Allows users to choose sorting options like by content, date, or type. - **Dry Run Mode:** Checks sorting results before committing changes. - **Silent Mode:** Logs all operations to a text file for quieter operation. ### Tech Stack: Python ### Why It Matters: This tool simplifies the organization of digital files, making it easier to find and manage your documents, while maintaining your privacy and security.

Overview: Lazy Predict is a Python AutoML toolkit designed for benchmarking classification, regression, and time-series forecasting models. It includes GPU acceleration, MLflow tracking, and automatic preprocessing, making it a versatile solution for developers. Key features: - Over 40 built-in machine learning models - Automatic model selection for classification, regression, and time series forecasting - 20+ forecasting models: statistical, ML, deep learning, and pretrained foundation models - Automatic seasonal period detection via ACF - Multiple categorical encoding strategies (OneHot, Ordinal, Target, Binary) - Built-in MLflow integration for experiment tracking - GPU acceleration: XGBoost, LightGBM, CatBoost, cuML (RAPIDS), LSTM/GRU, TimesFM - Support for Python 3.9 through 3.13 - Custom metric evaluation support - Configurable timeout and cross-validation - Intel Extension for Scikit-learn acceleration support Tech stack: Python Why it matters: Lazy Predict simplifies the process of building and evaluating machine learning models, making it accessible to developers without extensive coding skills. It automates preprocessing, model selection, and experiment tracking, reducing the time and effort required for model development and deployment.